Documentation Index
Fetch the complete documentation index at: https://arize-ax.mintlify.dev/docs/llms.txt
Use this file to discover all available pages before exploring further.
What’s New
April 30, 2021New Datasets Tab for Models
We’ve made it easy to view the data sent in across various datasets. You now have a centralized view of the datasets to help you choose and configure the appropriate baseline for your model.
New Config Tab in Models
We’ve streamlined the ability to set model baselines and configure alerts for monitors. From the Model Overview page, click on the Config tab to adjust the settings:- Set or change a model baseline
- Easily toggle on/off alerts for auto-drift monitors
- See and edit alert contacts for auto-drift monitors
Easier Documentation Access
April 16, 2021 Our product documentation is now accessible directly from the platform navigation, making it faster than ever to find answers to your questions!
Select Production as a model baseline
We now support setting historical production data as a model baseline. This enables you to compare current production performance against a prior time period in production.
- Detecting slow bleeds in model drift over time
- Analyzing model performance seasonality
- Understanding if an upstream change to a model’s inputs is impacting production performance
New Dashboards tab on Model Overview
A new Dashboards tab is now available on the Model Overview page (or as we like to think of it: your model overview launchpad 🚀). This addition makes it simpler to locate any dashboards associated with a model and create new dashboards as you’re checking in on a model.
Additional dimensions available for model health grid
Drift score for features, predictions, and actuals are now available for model health on the Model Overview page, allowing for a more comprehensive at-a-glance view of model drift across all dimensions.
Model Health Metrics Automatically Surfaced in the Overview Tab
April 2, 2021 Even without setting up a monitor, we now surface up automatic model health metrics in the model overview tab. Today we support surfacing up missing values for features. The platform will automatically surface an alert when a feature is missing values - e.g. when percent empty is greater than 1%. This makes it easier and faster to catch data quality issues when they crop up. Additional alerts for metrics such as drift are coming soon!
Establishing a Model Baseline
The ability to define a model baseline for model performance comparison is now generally available. You can set baselines from training/validation and different model versions.
Enhancements
April 30, 2021P1 and P99 now available for Model Health
We added P1 and P99 metrics to the Model Health grid so you can quickly determine if there are any features, predictions, actuals values out of range.
Navigate Back to Models from a Dashboard
We’ve added the ability to quickly surface what models are available on a dashboard and navigate into the model for further analysis.
Added Support for Single API
We added support allowing a single API call to accept log and bulk_log record type individually or together. Events are now zipped server-side, reducing latency of data populating in the platform. Additionally, the API makes event publishing more straightforward, allowing you to more easily drop the API call into different pipelines.- Quicker bulk set up of drift monitors across all features/predictions/actuals
-
Added support for:
- False negative density on feature performance heatmaps
- Dimension Stats quantiles
- Quantiles support for drift
- Improvements to histogram binning for drift monitors
UX enhancements to help streamline your workflow
April 2, 2021- Explainability tab now available under the Model Page for a more centralized view of all aspects of your model’s health.
- Monitor creation workflow now available under the Monitors Page to make it easier to add new monitors, wherever you happen to be in the platform
- Typeaheads for Categorical Values are now live in the platform to help speed up your input searches
Bug Fixes
April 2, 2021 The following issues were recently resolved in the platform:- Could not filter on Feature in the Heatmap Dashboard
- Could not add a Class to Feature Slices Performance Dashboard
- Could not add filters for Prediction Value for Explainability
In the News
April 30, 2021Arize Earns Spot on Forbes AI 50
Forbes published a list of America’s Most Promising AI Companies, featuring Arize AI among a diverse list of AI-focused businesses. Check out the full list of honorees along with a highlight on what AI Founders think about jobs in and after the pandemic.
Algorithmia and Arize Partner to Enable Better ML Observability for Enterprises
We’re excited to share that Arize AI and Algorithmia are partnered to help organizations deliver better models to production, maximize their performance, and minimize model risk. A technical integration tutorial is available here in our documentation. Read the announcement.
Machine Learning Ecosystem 101 Whitepaper
It can be hard staying up to date as the ML ecosystem continues growing at a rapid pace. We put together a comprehensive overview of the machine learning ecosystem to help to navigate recommendations and best practices when it comes to infrastructure, lifecycle, and workflow. Download the paper. April 16, 2021 Why Business Executives Should be Hip to ML Tools It’s imperative for business leaders to understand how the technology their organization builds and employs powers their organization. We published a simple primer on the fundamentals of the machine learning stack that every exec should know. Read the post.
The Only 3 ML Tools You Need
April 2, 2021 In this piece we discuss the three ML tools you need to make your team successful in applying machine learning in your product. Read the blog.
Playbook on How to Monitor Your Models
If you missed the latest issue of The ML Engineer Newsletter, our playbook on best practices when monitoring models is now available. Check it out here.
Tammy Le Joins Arize as VP of Marketing
We are excited to have Tammy Le join us as VP of Marketing! Check out our welcome blog post here!